zfs_znode.c revision 224231
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25/* Portions Copyright 2007 Jeremy Teo */ 26 27#ifdef _KERNEL 28#include <sys/types.h> 29#include <sys/param.h> 30#include <sys/time.h> 31#include <sys/systm.h> 32#include <sys/sysmacros.h> 33#include <sys/resource.h> 34#include <sys/mntent.h> 35#include <sys/u8_textprep.h> 36#include <sys/dsl_dataset.h> 37#include <sys/vfs.h> 38#include <sys/vnode.h> 39#include <sys/file.h> 40#include <sys/kmem.h> 41#include <sys/errno.h> 42#include <sys/unistd.h> 43#include <sys/atomic.h> 44#include <sys/zfs_dir.h> 45#include <sys/zfs_acl.h> 46#include <sys/zfs_ioctl.h> 47#include <sys/zfs_rlock.h> 48#include <sys/zfs_fuid.h> 49#include <sys/dnode.h> 50#include <sys/fs/zfs.h> 51#include <sys/kidmap.h> 52#endif /* _KERNEL */ 53 54#include <sys/dmu.h> 55#include <sys/refcount.h> 56#include <sys/stat.h> 57#include <sys/zap.h> 58#include <sys/zfs_znode.h> 59#include <sys/sa.h> 60#include <sys/zfs_sa.h> 61#include <sys/zfs_stat.h> 62#include <sys/refcount.h> 63 64#include "zfs_prop.h" 65#include "zfs_comutil.h" 66 67/* Used by fstat(1). */ 68SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 69 "sizeof(znode_t)"); 70 71/* 72 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 73 * turned on when DEBUG is also defined. 74 */ 75#ifdef DEBUG 76#define ZNODE_STATS 77#endif /* DEBUG */ 78 79#ifdef ZNODE_STATS 80#define ZNODE_STAT_ADD(stat) ((stat)++) 81#else 82#define ZNODE_STAT_ADD(stat) /* nothing */ 83#endif /* ZNODE_STATS */ 84 85/* 86 * Functions needed for userland (ie: libzpool) are not put under 87 * #ifdef_KERNEL; the rest of the functions have dependencies 88 * (such as VFS logic) that will not compile easily in userland. 89 */ 90#ifdef _KERNEL 91/* 92 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 93 * be freed before it can be safely accessed. 94 */ 95krwlock_t zfsvfs_lock; 96 97static kmem_cache_t *znode_cache = NULL; 98 99/*ARGSUSED*/ 100static void 101znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 102{ 103 /* 104 * We should never drop all dbuf refs without first clearing 105 * the eviction callback. 106 */ 107 panic("evicting znode %p\n", user_ptr); 108} 109 110extern struct vop_vector zfs_vnodeops; 111extern struct vop_vector zfs_fifoops; 112extern struct vop_vector zfs_shareops; 113 114/* 115 * XXX: We cannot use this function as a cache constructor, because 116 * there is one global cache for all file systems and we need 117 * to pass vfsp here, which is not possible, because argument 118 * 'cdrarg' is defined at kmem_cache_create() time. 119 */ 120/*ARGSUSED*/ 121static int 122zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 123{ 124 znode_t *zp = buf; 125 vnode_t *vp; 126 vfs_t *vfsp = arg; 127 int error; 128 129 POINTER_INVALIDATE(&zp->z_zfsvfs); 130 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 131 132 if (vfsp != NULL) { 133 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 134 if (error != 0 && (kmflags & KM_NOSLEEP)) 135 return (-1); 136 ASSERT(error == 0); 137 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 138 zp->z_vnode = vp; 139 vp->v_data = (caddr_t)zp; 140 VN_LOCK_AREC(vp); 141 VN_LOCK_ASHARE(vp); 142 } else { 143 zp->z_vnode = NULL; 144 } 145 146 list_link_init(&zp->z_link_node); 147 148 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 149 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 150 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 151 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 152 153 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 154 avl_create(&zp->z_range_avl, zfs_range_compare, 155 sizeof (rl_t), offsetof(rl_t, r_node)); 156 157 zp->z_dirlocks = NULL; 158 zp->z_acl_cached = NULL; 159 zp->z_moved = 0; 160 return (0); 161} 162 163/*ARGSUSED*/ 164static void 165zfs_znode_cache_destructor(void *buf, void *arg) 166{ 167 znode_t *zp = buf; 168 169 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 170 ASSERT(ZTOV(zp) == NULL); 171 vn_free(ZTOV(zp)); 172 ASSERT(!list_link_active(&zp->z_link_node)); 173 mutex_destroy(&zp->z_lock); 174 rw_destroy(&zp->z_parent_lock); 175 rw_destroy(&zp->z_name_lock); 176 mutex_destroy(&zp->z_acl_lock); 177 avl_destroy(&zp->z_range_avl); 178 mutex_destroy(&zp->z_range_lock); 179 180 ASSERT(zp->z_dirlocks == NULL); 181 ASSERT(zp->z_acl_cached == NULL); 182} 183 184#ifdef ZNODE_STATS 185static struct { 186 uint64_t zms_zfsvfs_invalid; 187 uint64_t zms_zfsvfs_recheck1; 188 uint64_t zms_zfsvfs_unmounted; 189 uint64_t zms_zfsvfs_recheck2; 190 uint64_t zms_obj_held; 191 uint64_t zms_vnode_locked; 192 uint64_t zms_not_only_dnlc; 193} znode_move_stats; 194#endif /* ZNODE_STATS */ 195 196#ifdef sun 197static void 198zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 199{ 200 vnode_t *vp; 201 202 /* Copy fields. */ 203 nzp->z_zfsvfs = ozp->z_zfsvfs; 204 205 /* Swap vnodes. */ 206 vp = nzp->z_vnode; 207 nzp->z_vnode = ozp->z_vnode; 208 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 209 ZTOV(ozp)->v_data = ozp; 210 ZTOV(nzp)->v_data = nzp; 211 212 nzp->z_id = ozp->z_id; 213 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 214 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 215 nzp->z_unlinked = ozp->z_unlinked; 216 nzp->z_atime_dirty = ozp->z_atime_dirty; 217 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 218 nzp->z_blksz = ozp->z_blksz; 219 nzp->z_seq = ozp->z_seq; 220 nzp->z_mapcnt = ozp->z_mapcnt; 221 nzp->z_gen = ozp->z_gen; 222 nzp->z_sync_cnt = ozp->z_sync_cnt; 223 nzp->z_is_sa = ozp->z_is_sa; 224 nzp->z_sa_hdl = ozp->z_sa_hdl; 225 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 226 nzp->z_links = ozp->z_links; 227 nzp->z_size = ozp->z_size; 228 nzp->z_pflags = ozp->z_pflags; 229 nzp->z_uid = ozp->z_uid; 230 nzp->z_gid = ozp->z_gid; 231 nzp->z_mode = ozp->z_mode; 232 233 /* 234 * Since this is just an idle znode and kmem is already dealing with 235 * memory pressure, release any cached ACL. 236 */ 237 if (ozp->z_acl_cached) { 238 zfs_acl_free(ozp->z_acl_cached); 239 ozp->z_acl_cached = NULL; 240 } 241 242 sa_set_userp(nzp->z_sa_hdl, nzp); 243 244 /* 245 * Invalidate the original znode by clearing fields that provide a 246 * pointer back to the znode. Set the low bit of the vfs pointer to 247 * ensure that zfs_znode_move() recognizes the znode as invalid in any 248 * subsequent callback. 249 */ 250 ozp->z_sa_hdl = NULL; 251 POINTER_INVALIDATE(&ozp->z_zfsvfs); 252 253 /* 254 * Mark the znode. 255 */ 256 nzp->z_moved = 1; 257 ozp->z_moved = (uint8_t)-1; 258} 259 260/*ARGSUSED*/ 261static kmem_cbrc_t 262zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 263{ 264 znode_t *ozp = buf, *nzp = newbuf; 265 zfsvfs_t *zfsvfs; 266 vnode_t *vp; 267 268 /* 269 * The znode is on the file system's list of known znodes if the vfs 270 * pointer is valid. We set the low bit of the vfs pointer when freeing 271 * the znode to invalidate it, and the memory patterns written by kmem 272 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 273 * created znode sets the vfs pointer last of all to indicate that the 274 * znode is known and in a valid state to be moved by this function. 275 */ 276 zfsvfs = ozp->z_zfsvfs; 277 if (!POINTER_IS_VALID(zfsvfs)) { 278 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 279 return (KMEM_CBRC_DONT_KNOW); 280 } 281 282 /* 283 * Close a small window in which it's possible that the filesystem could 284 * be unmounted and freed, and zfsvfs, though valid in the previous 285 * statement, could point to unrelated memory by the time we try to 286 * prevent the filesystem from being unmounted. 287 */ 288 rw_enter(&zfsvfs_lock, RW_WRITER); 289 if (zfsvfs != ozp->z_zfsvfs) { 290 rw_exit(&zfsvfs_lock); 291 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 292 return (KMEM_CBRC_DONT_KNOW); 293 } 294 295 /* 296 * If the znode is still valid, then so is the file system. We know that 297 * no valid file system can be freed while we hold zfsvfs_lock, so we 298 * can safely ensure that the filesystem is not and will not be 299 * unmounted. The next statement is equivalent to ZFS_ENTER(). 300 */ 301 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 302 if (zfsvfs->z_unmounted) { 303 ZFS_EXIT(zfsvfs); 304 rw_exit(&zfsvfs_lock); 305 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 306 return (KMEM_CBRC_DONT_KNOW); 307 } 308 rw_exit(&zfsvfs_lock); 309 310 mutex_enter(&zfsvfs->z_znodes_lock); 311 /* 312 * Recheck the vfs pointer in case the znode was removed just before 313 * acquiring the lock. 314 */ 315 if (zfsvfs != ozp->z_zfsvfs) { 316 mutex_exit(&zfsvfs->z_znodes_lock); 317 ZFS_EXIT(zfsvfs); 318 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 319 return (KMEM_CBRC_DONT_KNOW); 320 } 321 322 /* 323 * At this point we know that as long as we hold z_znodes_lock, the 324 * znode cannot be freed and fields within the znode can be safely 325 * accessed. Now, prevent a race with zfs_zget(). 326 */ 327 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 328 mutex_exit(&zfsvfs->z_znodes_lock); 329 ZFS_EXIT(zfsvfs); 330 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 331 return (KMEM_CBRC_LATER); 332 } 333 334 vp = ZTOV(ozp); 335 if (mutex_tryenter(&vp->v_lock) == 0) { 336 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 337 mutex_exit(&zfsvfs->z_znodes_lock); 338 ZFS_EXIT(zfsvfs); 339 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 340 return (KMEM_CBRC_LATER); 341 } 342 343 /* Only move znodes that are referenced _only_ by the DNLC. */ 344 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 345 mutex_exit(&vp->v_lock); 346 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 347 mutex_exit(&zfsvfs->z_znodes_lock); 348 ZFS_EXIT(zfsvfs); 349 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 350 return (KMEM_CBRC_LATER); 351 } 352 353 /* 354 * The znode is known and in a valid state to move. We're holding the 355 * locks needed to execute the critical section. 356 */ 357 zfs_znode_move_impl(ozp, nzp); 358 mutex_exit(&vp->v_lock); 359 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 360 361 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 362 mutex_exit(&zfsvfs->z_znodes_lock); 363 ZFS_EXIT(zfsvfs); 364 365 return (KMEM_CBRC_YES); 366} 367#endif /* sun */ 368 369void 370zfs_znode_init(void) 371{ 372 /* 373 * Initialize zcache 374 */ 375 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 376 ASSERT(znode_cache == NULL); 377 znode_cache = kmem_cache_create("zfs_znode_cache", 378 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 379 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 380 kmem_cache_set_move(znode_cache, zfs_znode_move); 381} 382 383void 384zfs_znode_fini(void) 385{ 386#ifdef sun 387 /* 388 * Cleanup vfs & vnode ops 389 */ 390 zfs_remove_op_tables(); 391#endif /* sun */ 392 393 /* 394 * Cleanup zcache 395 */ 396 if (znode_cache) 397 kmem_cache_destroy(znode_cache); 398 znode_cache = NULL; 399 rw_destroy(&zfsvfs_lock); 400} 401 402#ifdef sun 403struct vnodeops *zfs_dvnodeops; 404struct vnodeops *zfs_fvnodeops; 405struct vnodeops *zfs_symvnodeops; 406struct vnodeops *zfs_xdvnodeops; 407struct vnodeops *zfs_evnodeops; 408struct vnodeops *zfs_sharevnodeops; 409 410void 411zfs_remove_op_tables() 412{ 413 /* 414 * Remove vfs ops 415 */ 416 ASSERT(zfsfstype); 417 (void) vfs_freevfsops_by_type(zfsfstype); 418 zfsfstype = 0; 419 420 /* 421 * Remove vnode ops 422 */ 423 if (zfs_dvnodeops) 424 vn_freevnodeops(zfs_dvnodeops); 425 if (zfs_fvnodeops) 426 vn_freevnodeops(zfs_fvnodeops); 427 if (zfs_symvnodeops) 428 vn_freevnodeops(zfs_symvnodeops); 429 if (zfs_xdvnodeops) 430 vn_freevnodeops(zfs_xdvnodeops); 431 if (zfs_evnodeops) 432 vn_freevnodeops(zfs_evnodeops); 433 if (zfs_sharevnodeops) 434 vn_freevnodeops(zfs_sharevnodeops); 435 436 zfs_dvnodeops = NULL; 437 zfs_fvnodeops = NULL; 438 zfs_symvnodeops = NULL; 439 zfs_xdvnodeops = NULL; 440 zfs_evnodeops = NULL; 441 zfs_sharevnodeops = NULL; 442} 443 444extern const fs_operation_def_t zfs_dvnodeops_template[]; 445extern const fs_operation_def_t zfs_fvnodeops_template[]; 446extern const fs_operation_def_t zfs_xdvnodeops_template[]; 447extern const fs_operation_def_t zfs_symvnodeops_template[]; 448extern const fs_operation_def_t zfs_evnodeops_template[]; 449extern const fs_operation_def_t zfs_sharevnodeops_template[]; 450 451int 452zfs_create_op_tables() 453{ 454 int error; 455 456 /* 457 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 458 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 459 * In this case we just return as the ops vectors are already set up. 460 */ 461 if (zfs_dvnodeops) 462 return (0); 463 464 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 465 &zfs_dvnodeops); 466 if (error) 467 return (error); 468 469 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 470 &zfs_fvnodeops); 471 if (error) 472 return (error); 473 474 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 475 &zfs_symvnodeops); 476 if (error) 477 return (error); 478 479 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 480 &zfs_xdvnodeops); 481 if (error) 482 return (error); 483 484 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 485 &zfs_evnodeops); 486 if (error) 487 return (error); 488 489 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 490 &zfs_sharevnodeops); 491 492 return (error); 493} 494#endif /* sun */ 495 496int 497zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 498{ 499 zfs_acl_ids_t acl_ids; 500 vattr_t vattr; 501 znode_t *sharezp; 502 vnode_t *vp, vnode; 503 znode_t *zp; 504 int error; 505 506 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 507 vattr.va_type = VDIR; 508 vattr.va_mode = S_IFDIR|0555; 509 vattr.va_uid = crgetuid(kcred); 510 vattr.va_gid = crgetgid(kcred); 511 512 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 513 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 514 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 515 sharezp->z_moved = 0; 516 sharezp->z_unlinked = 0; 517 sharezp->z_atime_dirty = 0; 518 sharezp->z_zfsvfs = zfsvfs; 519 sharezp->z_is_sa = zfsvfs->z_use_sa; 520 521 sharezp->z_vnode = &vnode; 522 vnode.v_data = sharezp; 523 524 vp = ZTOV(sharezp); 525 vp->v_type = VDIR; 526 527 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 528 kcred, NULL, &acl_ids)); 529 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 530 ASSERT3P(zp, ==, sharezp); 531 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 532 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 533 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 534 zfsvfs->z_shares_dir = sharezp->z_id; 535 536 zfs_acl_ids_free(&acl_ids); 537 ZTOV(sharezp)->v_data = NULL; 538 ZTOV(sharezp)->v_count = 0; 539 ZTOV(sharezp)->v_holdcnt = 0; 540 zp->z_vnode = NULL; 541 sa_handle_destroy(sharezp->z_sa_hdl); 542 sharezp->z_vnode = NULL; 543 kmem_cache_free(znode_cache, sharezp); 544 545 return (error); 546} 547 548/* 549 * define a couple of values we need available 550 * for both 64 and 32 bit environments. 551 */ 552#ifndef NBITSMINOR64 553#define NBITSMINOR64 32 554#endif 555#ifndef MAXMAJ64 556#define MAXMAJ64 0xffffffffUL 557#endif 558#ifndef MAXMIN64 559#define MAXMIN64 0xffffffffUL 560#endif 561 562/* 563 * Create special expldev for ZFS private use. 564 * Can't use standard expldev since it doesn't do 565 * what we want. The standard expldev() takes a 566 * dev32_t in LP64 and expands it to a long dev_t. 567 * We need an interface that takes a dev32_t in ILP32 568 * and expands it to a long dev_t. 569 */ 570static uint64_t 571zfs_expldev(dev_t dev) 572{ 573 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 574} 575/* 576 * Special cmpldev for ZFS private use. 577 * Can't use standard cmpldev since it takes 578 * a long dev_t and compresses it to dev32_t in 579 * LP64. We need to do a compaction of a long dev_t 580 * to a dev32_t in ILP32. 581 */ 582dev_t 583zfs_cmpldev(uint64_t dev) 584{ 585 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 586} 587 588static void 589zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 590 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 591{ 592 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 593 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 594 595 mutex_enter(&zp->z_lock); 596 597 ASSERT(zp->z_sa_hdl == NULL); 598 ASSERT(zp->z_acl_cached == NULL); 599 if (sa_hdl == NULL) { 600 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 601 SA_HDL_SHARED, &zp->z_sa_hdl)); 602 } else { 603 zp->z_sa_hdl = sa_hdl; 604 sa_set_userp(sa_hdl, zp); 605 } 606 607 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 608 609 /* 610 * Slap on VROOT if we are the root znode 611 */ 612 if (zp->z_id == zfsvfs->z_root) 613 ZTOV(zp)->v_flag |= VROOT; 614 615 mutex_exit(&zp->z_lock); 616 vn_exists(ZTOV(zp)); 617} 618 619void 620zfs_znode_dmu_fini(znode_t *zp) 621{ 622 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 623 zp->z_unlinked || 624 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 625 626 sa_handle_destroy(zp->z_sa_hdl); 627 zp->z_sa_hdl = NULL; 628} 629 630/* 631 * Construct a new znode/vnode and intialize. 632 * 633 * This does not do a call to dmu_set_user() that is 634 * up to the caller to do, in case you don't want to 635 * return the znode 636 */ 637static znode_t * 638zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 639 dmu_object_type_t obj_type, sa_handle_t *hdl) 640{ 641 znode_t *zp; 642 vnode_t *vp; 643 uint64_t mode; 644 uint64_t parent; 645 sa_bulk_attr_t bulk[9]; 646 int count = 0; 647 648 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 649 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 650 651 ASSERT(zp->z_dirlocks == NULL); 652 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 653 zp->z_moved = 0; 654 655 /* 656 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 657 * the zfs_znode_move() callback. 658 */ 659 zp->z_sa_hdl = NULL; 660 zp->z_unlinked = 0; 661 zp->z_atime_dirty = 0; 662 zp->z_mapcnt = 0; 663 zp->z_id = db->db_object; 664 zp->z_blksz = blksz; 665 zp->z_seq = 0x7A4653; 666 zp->z_sync_cnt = 0; 667 668 vp = ZTOV(zp); 669 670 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 671 672 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 673 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 674 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 675 &zp->z_size, 8); 676 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 677 &zp->z_links, 8); 678 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 679 &zp->z_pflags, 8); 680 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 681 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 682 &zp->z_atime, 16); 683 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 684 &zp->z_uid, 8); 685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 686 &zp->z_gid, 8); 687 688 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 689 if (hdl == NULL) 690 sa_handle_destroy(zp->z_sa_hdl); 691 kmem_cache_free(znode_cache, zp); 692 return (NULL); 693 } 694 695 zp->z_mode = mode; 696 697 vp->v_type = IFTOVT((mode_t)mode); 698 699 switch (vp->v_type) { 700 case VDIR: 701 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 702 break; 703 case VBLK: 704 case VCHR: 705 { 706 uint64_t rdev; 707 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 708 &rdev, sizeof (rdev)) == 0); 709 710 zp->z_rdev = zfs_cmpldev(rdev); 711 } 712 break; 713 case VFIFO: 714 vp->v_op = &zfs_fifoops; 715 break; 716 case VREG: 717 if (parent == zfsvfs->z_shares_dir) { 718 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 719 vp->v_op = &zfs_shareops; 720 } 721 break; 722 } 723 if (vp->v_type != VFIFO) 724 VN_LOCK_ASHARE(vp); 725 726 mutex_enter(&zfsvfs->z_znodes_lock); 727 list_insert_tail(&zfsvfs->z_all_znodes, zp); 728 membar_producer(); 729 /* 730 * Everything else must be valid before assigning z_zfsvfs makes the 731 * znode eligible for zfs_znode_move(). 732 */ 733 zp->z_zfsvfs = zfsvfs; 734 mutex_exit(&zfsvfs->z_znodes_lock); 735 736 VFS_HOLD(zfsvfs->z_vfs); 737 return (zp); 738} 739 740static uint64_t empty_xattr; 741static uint64_t pad[4]; 742static zfs_acl_phys_t acl_phys; 743/* 744 * Create a new DMU object to hold a zfs znode. 745 * 746 * IN: dzp - parent directory for new znode 747 * vap - file attributes for new znode 748 * tx - dmu transaction id for zap operations 749 * cr - credentials of caller 750 * flag - flags: 751 * IS_ROOT_NODE - new object will be root 752 * IS_XATTR - new object is an attribute 753 * bonuslen - length of bonus buffer 754 * setaclp - File/Dir initial ACL 755 * fuidp - Tracks fuid allocation. 756 * 757 * OUT: zpp - allocated znode 758 * 759 */ 760void 761zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 762 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 763{ 764 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 765 uint64_t mode, size, links, parent, pflags; 766 uint64_t dzp_pflags = 0; 767 uint64_t rdev = 0; 768 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 769 dmu_buf_t *db; 770 timestruc_t now; 771 uint64_t gen, obj; 772 int err; 773 int bonuslen; 774 sa_handle_t *sa_hdl; 775 dmu_object_type_t obj_type; 776 sa_bulk_attr_t sa_attrs[ZPL_END]; 777 int cnt = 0; 778 zfs_acl_locator_cb_t locate = { 0 }; 779 780 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 781 782 if (zfsvfs->z_replay) { 783 obj = vap->va_nodeid; 784 now = vap->va_ctime; /* see zfs_replay_create() */ 785 gen = vap->va_nblocks; /* ditto */ 786 } else { 787 obj = 0; 788 gethrestime(&now); 789 gen = dmu_tx_get_txg(tx); 790 } 791 792 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 793 bonuslen = (obj_type == DMU_OT_SA) ? 794 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 795 796 /* 797 * Create a new DMU object. 798 */ 799 /* 800 * There's currently no mechanism for pre-reading the blocks that will 801 * be needed to allocate a new object, so we accept the small chance 802 * that there will be an i/o error and we will fail one of the 803 * assertions below. 804 */ 805 if (vap->va_type == VDIR) { 806 if (zfsvfs->z_replay) { 807 err = zap_create_claim_norm(zfsvfs->z_os, obj, 808 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 809 obj_type, bonuslen, tx); 810 ASSERT3U(err, ==, 0); 811 } else { 812 obj = zap_create_norm(zfsvfs->z_os, 813 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 814 obj_type, bonuslen, tx); 815 } 816 } else { 817 if (zfsvfs->z_replay) { 818 err = dmu_object_claim(zfsvfs->z_os, obj, 819 DMU_OT_PLAIN_FILE_CONTENTS, 0, 820 obj_type, bonuslen, tx); 821 ASSERT3U(err, ==, 0); 822 } else { 823 obj = dmu_object_alloc(zfsvfs->z_os, 824 DMU_OT_PLAIN_FILE_CONTENTS, 0, 825 obj_type, bonuslen, tx); 826 } 827 } 828 829 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 830 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 831 832 /* 833 * If this is the root, fix up the half-initialized parent pointer 834 * to reference the just-allocated physical data area. 835 */ 836 if (flag & IS_ROOT_NODE) { 837 dzp->z_id = obj; 838 } else { 839 dzp_pflags = dzp->z_pflags; 840 } 841 842 /* 843 * If parent is an xattr, so am I. 844 */ 845 if (dzp_pflags & ZFS_XATTR) { 846 flag |= IS_XATTR; 847 } 848 849 if (zfsvfs->z_use_fuids) 850 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 851 else 852 pflags = 0; 853 854 if (vap->va_type == VDIR) { 855 size = 2; /* contents ("." and "..") */ 856 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 857 } else { 858 size = links = 0; 859 } 860 861 if (vap->va_type == VBLK || vap->va_type == VCHR) { 862 rdev = zfs_expldev(vap->va_rdev); 863 } 864 865 parent = dzp->z_id; 866 mode = acl_ids->z_mode; 867 if (flag & IS_XATTR) 868 pflags |= ZFS_XATTR; 869 870 /* 871 * No execs denied will be deterimed when zfs_mode_compute() is called. 872 */ 873 pflags |= acl_ids->z_aclp->z_hints & 874 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 875 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 876 877 ZFS_TIME_ENCODE(&now, crtime); 878 ZFS_TIME_ENCODE(&now, ctime); 879 880 if (vap->va_mask & AT_ATIME) { 881 ZFS_TIME_ENCODE(&vap->va_atime, atime); 882 } else { 883 ZFS_TIME_ENCODE(&now, atime); 884 } 885 886 if (vap->va_mask & AT_MTIME) { 887 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 888 } else { 889 ZFS_TIME_ENCODE(&now, mtime); 890 } 891 892 /* Now add in all of the "SA" attributes */ 893 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 894 &sa_hdl)); 895 896 /* 897 * Setup the array of attributes to be replaced/set on the new file 898 * 899 * order for DMU_OT_ZNODE is critical since it needs to be constructed 900 * in the old znode_phys_t format. Don't change this ordering 901 */ 902 903 if (obj_type == DMU_OT_ZNODE) { 904 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 905 NULL, &atime, 16); 906 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 907 NULL, &mtime, 16); 908 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 909 NULL, &ctime, 16); 910 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 911 NULL, &crtime, 16); 912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 913 NULL, &gen, 8); 914 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 915 NULL, &mode, 8); 916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 917 NULL, &size, 8); 918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 919 NULL, &parent, 8); 920 } else { 921 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 922 NULL, &mode, 8); 923 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 924 NULL, &size, 8); 925 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 926 NULL, &gen, 8); 927 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 928 &acl_ids->z_fuid, 8); 929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 930 &acl_ids->z_fgid, 8); 931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 932 NULL, &parent, 8); 933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 934 NULL, &pflags, 8); 935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 936 NULL, &atime, 16); 937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 938 NULL, &mtime, 16); 939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 940 NULL, &ctime, 16); 941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 942 NULL, &crtime, 16); 943 } 944 945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 946 947 if (obj_type == DMU_OT_ZNODE) { 948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 949 &empty_xattr, 8); 950 } 951 if (obj_type == DMU_OT_ZNODE || 952 (vap->va_type == VBLK || vap->va_type == VCHR)) { 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 954 NULL, &rdev, 8); 955 956 } 957 if (obj_type == DMU_OT_ZNODE) { 958 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 959 NULL, &pflags, 8); 960 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 961 &acl_ids->z_fuid, 8); 962 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 963 &acl_ids->z_fgid, 8); 964 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 965 sizeof (uint64_t) * 4); 966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 967 &acl_phys, sizeof (zfs_acl_phys_t)); 968 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 970 &acl_ids->z_aclp->z_acl_count, 8); 971 locate.cb_aclp = acl_ids->z_aclp; 972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 973 zfs_acl_data_locator, &locate, 974 acl_ids->z_aclp->z_acl_bytes); 975 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 976 acl_ids->z_fuid, acl_ids->z_fgid); 977 } 978 979 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 980 981 if (!(flag & IS_ROOT_NODE)) { 982 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 983 ASSERT(*zpp != NULL); 984 } else { 985 /* 986 * If we are creating the root node, the "parent" we 987 * passed in is the znode for the root. 988 */ 989 *zpp = dzp; 990 991 (*zpp)->z_sa_hdl = sa_hdl; 992 } 993 994 (*zpp)->z_pflags = pflags; 995 (*zpp)->z_mode = mode; 996 997 if (vap->va_mask & AT_XVATTR) 998 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 999 1000 if (obj_type == DMU_OT_ZNODE || 1001 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1002 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1003 ASSERT3P(err, ==, 0); 1004 } 1005 if (!(flag & IS_ROOT_NODE)) { 1006 vnode_t *vp; 1007 1008 vp = ZTOV(*zpp); 1009 vp->v_vflag |= VV_FORCEINSMQ; 1010 err = insmntque(vp, zfsvfs->z_vfs); 1011 vp->v_vflag &= ~VV_FORCEINSMQ; 1012 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1013 } 1014 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1015} 1016 1017/* 1018 * zfs_xvattr_set only updates the in-core attributes 1019 * it is assumed the caller will be doing an sa_bulk_update 1020 * to push the changes out 1021 */ 1022void 1023zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1024{ 1025 xoptattr_t *xoap; 1026 1027 xoap = xva_getxoptattr(xvap); 1028 ASSERT(xoap); 1029 1030 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1031 uint64_t times[2]; 1032 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1033 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1034 ×, sizeof (times), tx); 1035 XVA_SET_RTN(xvap, XAT_CREATETIME); 1036 } 1037 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1038 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1039 zp->z_pflags, tx); 1040 XVA_SET_RTN(xvap, XAT_READONLY); 1041 } 1042 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1043 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1044 zp->z_pflags, tx); 1045 XVA_SET_RTN(xvap, XAT_HIDDEN); 1046 } 1047 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1048 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1049 zp->z_pflags, tx); 1050 XVA_SET_RTN(xvap, XAT_SYSTEM); 1051 } 1052 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1053 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1054 zp->z_pflags, tx); 1055 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1056 } 1057 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1058 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1059 zp->z_pflags, tx); 1060 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1061 } 1062 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1063 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1064 zp->z_pflags, tx); 1065 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1066 } 1067 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1068 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1069 zp->z_pflags, tx); 1070 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1071 } 1072 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1073 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1074 zp->z_pflags, tx); 1075 XVA_SET_RTN(xvap, XAT_NODUMP); 1076 } 1077 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1078 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1079 zp->z_pflags, tx); 1080 XVA_SET_RTN(xvap, XAT_OPAQUE); 1081 } 1082 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1083 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1084 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1085 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1086 } 1087 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1088 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1089 zp->z_pflags, tx); 1090 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1091 } 1092 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1093 zfs_sa_set_scanstamp(zp, xvap, tx); 1094 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1095 } 1096 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1097 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1098 zp->z_pflags, tx); 1099 XVA_SET_RTN(xvap, XAT_REPARSE); 1100 } 1101 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1102 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1103 zp->z_pflags, tx); 1104 XVA_SET_RTN(xvap, XAT_OFFLINE); 1105 } 1106 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1107 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1108 zp->z_pflags, tx); 1109 XVA_SET_RTN(xvap, XAT_SPARSE); 1110 } 1111} 1112 1113int 1114zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1115{ 1116 dmu_object_info_t doi; 1117 dmu_buf_t *db; 1118 znode_t *zp; 1119 int err; 1120 sa_handle_t *hdl; 1121 int first = 1; 1122 1123 *zpp = NULL; 1124 1125again: 1126 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1127 1128 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1129 if (err) { 1130 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1131 return (err); 1132 } 1133 1134 dmu_object_info_from_db(db, &doi); 1135 if (doi.doi_bonus_type != DMU_OT_SA && 1136 (doi.doi_bonus_type != DMU_OT_ZNODE || 1137 (doi.doi_bonus_type == DMU_OT_ZNODE && 1138 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1139 sa_buf_rele(db, NULL); 1140 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1141 return (EINVAL); 1142 } 1143 1144 hdl = dmu_buf_get_user(db); 1145 if (hdl != NULL) { 1146 zp = sa_get_userdata(hdl); 1147 1148 1149 /* 1150 * Since "SA" does immediate eviction we 1151 * should never find a sa handle that doesn't 1152 * know about the znode. 1153 */ 1154 1155 ASSERT3P(zp, !=, NULL); 1156 1157 mutex_enter(&zp->z_lock); 1158 ASSERT3U(zp->z_id, ==, obj_num); 1159 if (zp->z_unlinked) { 1160 err = ENOENT; 1161 } else { 1162 vnode_t *vp; 1163 int dying = 0; 1164 1165 vp = ZTOV(zp); 1166 if (vp == NULL) 1167 dying = 1; 1168 else { 1169 VN_HOLD(vp); 1170 if ((vp->v_iflag & VI_DOOMED) != 0) { 1171 dying = 1; 1172 /* 1173 * Don't VN_RELE() vnode here, because 1174 * it can call vn_lock() which creates 1175 * LOR between vnode lock and znode 1176 * lock. We will VN_RELE() the vnode 1177 * after droping znode lock. 1178 */ 1179 } 1180 } 1181 if (dying) { 1182 if (first) { 1183 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1184 first = 0; 1185 } 1186 /* 1187 * znode is dying so we can't reuse it, we must 1188 * wait until destruction is completed. 1189 */ 1190 sa_buf_rele(db, NULL); 1191 mutex_exit(&zp->z_lock); 1192 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1193 if (vp != NULL) 1194 VN_RELE(vp); 1195 tsleep(zp, 0, "zcollide", 1); 1196 goto again; 1197 } 1198 *zpp = zp; 1199 err = 0; 1200 } 1201 sa_buf_rele(db, NULL); 1202 mutex_exit(&zp->z_lock); 1203 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1204 return (err); 1205 } 1206 1207 /* 1208 * Not found create new znode/vnode 1209 * but only if file exists. 1210 * 1211 * There is a small window where zfs_vget() could 1212 * find this object while a file create is still in 1213 * progress. This is checked for in zfs_znode_alloc() 1214 * 1215 * if zfs_znode_alloc() fails it will drop the hold on the 1216 * bonus buffer. 1217 */ 1218 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1219 doi.doi_bonus_type, NULL); 1220 if (zp == NULL) { 1221 err = ENOENT; 1222 } else { 1223 *zpp = zp; 1224 } 1225 if (err == 0) { 1226 vnode_t *vp = ZTOV(zp); 1227 1228 err = insmntque(vp, zfsvfs->z_vfs); 1229 if (err == 0) 1230 VOP_UNLOCK(vp, 0); 1231 else { 1232 zp->z_vnode = NULL; 1233 zfs_znode_dmu_fini(zp); 1234 zfs_znode_free(zp); 1235 *zpp = NULL; 1236 } 1237 } 1238 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1239 return (err); 1240} 1241 1242int 1243zfs_rezget(znode_t *zp) 1244{ 1245 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1246 dmu_object_info_t doi; 1247 dmu_buf_t *db; 1248 uint64_t obj_num = zp->z_id; 1249 uint64_t mode, size; 1250 sa_bulk_attr_t bulk[8]; 1251 int err; 1252 int count = 0; 1253 uint64_t gen; 1254 1255 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1256 1257 mutex_enter(&zp->z_acl_lock); 1258 if (zp->z_acl_cached) { 1259 zfs_acl_free(zp->z_acl_cached); 1260 zp->z_acl_cached = NULL; 1261 } 1262 1263 mutex_exit(&zp->z_acl_lock); 1264 ASSERT(zp->z_sa_hdl == NULL); 1265 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1266 if (err) { 1267 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1268 return (err); 1269 } 1270 1271 dmu_object_info_from_db(db, &doi); 1272 if (doi.doi_bonus_type != DMU_OT_SA && 1273 (doi.doi_bonus_type != DMU_OT_ZNODE || 1274 (doi.doi_bonus_type == DMU_OT_ZNODE && 1275 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1276 sa_buf_rele(db, NULL); 1277 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1278 return (EINVAL); 1279 } 1280 1281 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1282 size = zp->z_size; 1283 1284 /* reload cached values */ 1285 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1286 &gen, sizeof (gen)); 1287 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1288 &zp->z_size, sizeof (zp->z_size)); 1289 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1290 &zp->z_links, sizeof (zp->z_links)); 1291 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1292 &zp->z_pflags, sizeof (zp->z_pflags)); 1293 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1294 &zp->z_atime, sizeof (zp->z_atime)); 1295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1296 &zp->z_uid, sizeof (zp->z_uid)); 1297 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1298 &zp->z_gid, sizeof (zp->z_gid)); 1299 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1300 &mode, sizeof (mode)); 1301 1302 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1303 zfs_znode_dmu_fini(zp); 1304 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1305 return (EIO); 1306 } 1307 1308 zp->z_mode = mode; 1309 1310 if (gen != zp->z_gen) { 1311 zfs_znode_dmu_fini(zp); 1312 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1313 return (EIO); 1314 } 1315 1316 /* 1317 * XXXPJD: Not sure how is that possible, but under heavy 1318 * zfs recv -F load it happens that z_gen is the same, but 1319 * vnode type is different than znode type. This would mean 1320 * that for example regular file was replaced with directory 1321 * which has the same object number. 1322 */ 1323 if (ZTOV(zp) != NULL && 1324 ZTOV(zp)->v_type != IFTOVT((mode_t)zp->z_mode)) { 1325 zfs_znode_dmu_fini(zp); 1326 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1327 return (EIO); 1328 } 1329 1330 zp->z_unlinked = (zp->z_links == 0); 1331 zp->z_blksz = doi.doi_data_block_size; 1332 if (zp->z_size != size && ZTOV(zp) != NULL) 1333 vnode_pager_setsize(ZTOV(zp), zp->z_size); 1334 1335 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1336 1337 return (0); 1338} 1339 1340void 1341zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1342{ 1343 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1344 objset_t *os = zfsvfs->z_os; 1345 uint64_t obj = zp->z_id; 1346 uint64_t acl_obj = zfs_external_acl(zp); 1347 1348 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1349 if (acl_obj) { 1350 VERIFY(!zp->z_is_sa); 1351 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1352 } 1353 VERIFY(0 == dmu_object_free(os, obj, tx)); 1354 zfs_znode_dmu_fini(zp); 1355 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1356 zfs_znode_free(zp); 1357} 1358 1359void 1360zfs_zinactive(znode_t *zp) 1361{ 1362 vnode_t *vp = ZTOV(zp); 1363 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1364 uint64_t z_id = zp->z_id; 1365 int vfslocked; 1366 1367 ASSERT(zp->z_sa_hdl); 1368 1369 /* 1370 * Don't allow a zfs_zget() while were trying to release this znode 1371 */ 1372 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1373 1374 mutex_enter(&zp->z_lock); 1375 VI_LOCK(vp); 1376 if (vp->v_count > 0) { 1377 /* 1378 * If the hold count is greater than zero, somebody has 1379 * obtained a new reference on this znode while we were 1380 * processing it here, so we are done. 1381 */ 1382 VI_UNLOCK(vp); 1383 mutex_exit(&zp->z_lock); 1384 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1385 return; 1386 } 1387 VI_UNLOCK(vp); 1388 1389 /* 1390 * If this was the last reference to a file with no links, 1391 * remove the file from the file system. 1392 */ 1393 if (zp->z_unlinked) { 1394 mutex_exit(&zp->z_lock); 1395 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1396 ASSERT(vp->v_count == 0); 1397 vrecycle(vp, curthread); 1398 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1399 zfs_rmnode(zp); 1400 VFS_UNLOCK_GIANT(vfslocked); 1401 return; 1402 } 1403 1404 mutex_exit(&zp->z_lock); 1405 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1406} 1407 1408void 1409zfs_znode_free(znode_t *zp) 1410{ 1411 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1412 1413 ASSERT(ZTOV(zp) == NULL); 1414 ASSERT(zp->z_sa_hdl == NULL); 1415 mutex_enter(&zfsvfs->z_znodes_lock); 1416 POINTER_INVALIDATE(&zp->z_zfsvfs); 1417 list_remove(&zfsvfs->z_all_znodes, zp); 1418 mutex_exit(&zfsvfs->z_znodes_lock); 1419 1420 if (zp->z_acl_cached) { 1421 zfs_acl_free(zp->z_acl_cached); 1422 zp->z_acl_cached = NULL; 1423 } 1424 1425 kmem_cache_free(znode_cache, zp); 1426 1427 VFS_RELE(zfsvfs->z_vfs); 1428} 1429 1430void 1431zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1432 uint64_t ctime[2], boolean_t have_tx) 1433{ 1434 timestruc_t now; 1435 1436 gethrestime(&now); 1437 1438 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1439 zp->z_atime_dirty = 0; 1440 zp->z_seq++; 1441 } else { 1442 zp->z_atime_dirty = 1; 1443 } 1444 1445 if (flag & AT_ATIME) { 1446 ZFS_TIME_ENCODE(&now, zp->z_atime); 1447 } 1448 1449 if (flag & AT_MTIME) { 1450 ZFS_TIME_ENCODE(&now, mtime); 1451 if (zp->z_zfsvfs->z_use_fuids) { 1452 zp->z_pflags |= (ZFS_ARCHIVE | 1453 ZFS_AV_MODIFIED); 1454 } 1455 } 1456 1457 if (flag & AT_CTIME) { 1458 ZFS_TIME_ENCODE(&now, ctime); 1459 if (zp->z_zfsvfs->z_use_fuids) 1460 zp->z_pflags |= ZFS_ARCHIVE; 1461 } 1462} 1463 1464/* 1465 * Grow the block size for a file. 1466 * 1467 * IN: zp - znode of file to free data in. 1468 * size - requested block size 1469 * tx - open transaction. 1470 * 1471 * NOTE: this function assumes that the znode is write locked. 1472 */ 1473void 1474zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1475{ 1476 int error; 1477 u_longlong_t dummy; 1478 1479 if (size <= zp->z_blksz) 1480 return; 1481 /* 1482 * If the file size is already greater than the current blocksize, 1483 * we will not grow. If there is more than one block in a file, 1484 * the blocksize cannot change. 1485 */ 1486 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1487 return; 1488 1489 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1490 size, 0, tx); 1491 1492 if (error == ENOTSUP) 1493 return; 1494 ASSERT3U(error, ==, 0); 1495 1496 /* What blocksize did we actually get? */ 1497 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1498} 1499 1500#ifdef sun 1501/* 1502 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1503 * be calling back into the fs for a putpage(). E.g.: when truncating 1504 * a file, the pages being "thrown away* don't need to be written out. 1505 */ 1506/* ARGSUSED */ 1507static int 1508zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1509 int flags, cred_t *cr) 1510{ 1511 ASSERT(0); 1512 return (0); 1513} 1514#endif /* sun */ 1515 1516/* 1517 * Increase the file length 1518 * 1519 * IN: zp - znode of file to free data in. 1520 * end - new end-of-file 1521 * 1522 * RETURN: 0 if success 1523 * error code if failure 1524 */ 1525static int 1526zfs_extend(znode_t *zp, uint64_t end) 1527{ 1528 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1529 dmu_tx_t *tx; 1530 rl_t *rl; 1531 uint64_t newblksz; 1532 int error; 1533 1534 /* 1535 * We will change zp_size, lock the whole file. 1536 */ 1537 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1538 1539 /* 1540 * Nothing to do if file already at desired length. 1541 */ 1542 if (end <= zp->z_size) { 1543 zfs_range_unlock(rl); 1544 return (0); 1545 } 1546top: 1547 tx = dmu_tx_create(zfsvfs->z_os); 1548 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1549 zfs_sa_upgrade_txholds(tx, zp); 1550 if (end > zp->z_blksz && 1551 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1552 /* 1553 * We are growing the file past the current block size. 1554 */ 1555 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1556 ASSERT(!ISP2(zp->z_blksz)); 1557 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1558 } else { 1559 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1560 } 1561 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1562 } else { 1563 newblksz = 0; 1564 } 1565 1566 error = dmu_tx_assign(tx, TXG_NOWAIT); 1567 if (error) { 1568 if (error == ERESTART) { 1569 dmu_tx_wait(tx); 1570 dmu_tx_abort(tx); 1571 goto top; 1572 } 1573 dmu_tx_abort(tx); 1574 zfs_range_unlock(rl); 1575 return (error); 1576 } 1577 1578 if (newblksz) 1579 zfs_grow_blocksize(zp, newblksz, tx); 1580 1581 zp->z_size = end; 1582 1583 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1584 &zp->z_size, sizeof (zp->z_size), tx)); 1585 1586 vnode_pager_setsize(ZTOV(zp), end); 1587 1588 zfs_range_unlock(rl); 1589 1590 dmu_tx_commit(tx); 1591 1592 return (0); 1593} 1594 1595/* 1596 * Free space in a file. 1597 * 1598 * IN: zp - znode of file to free data in. 1599 * off - start of section to free. 1600 * len - length of section to free. 1601 * 1602 * RETURN: 0 if success 1603 * error code if failure 1604 */ 1605static int 1606zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1607{ 1608 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1609 rl_t *rl; 1610 int error; 1611 1612 /* 1613 * Lock the range being freed. 1614 */ 1615 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1616 1617 /* 1618 * Nothing to do if file already at desired length. 1619 */ 1620 if (off >= zp->z_size) { 1621 zfs_range_unlock(rl); 1622 return (0); 1623 } 1624 1625 if (off + len > zp->z_size) 1626 len = zp->z_size - off; 1627 1628 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1629 1630 if (error == 0) { 1631 /* 1632 * In FreeBSD we cannot free block in the middle of a file, 1633 * but only at the end of a file, so this code path should 1634 * never happen. 1635 */ 1636 vnode_pager_setsize(ZTOV(zp), off); 1637 } 1638 1639 zfs_range_unlock(rl); 1640 1641 return (error); 1642} 1643 1644/* 1645 * Truncate a file 1646 * 1647 * IN: zp - znode of file to free data in. 1648 * end - new end-of-file. 1649 * 1650 * RETURN: 0 if success 1651 * error code if failure 1652 */ 1653static int 1654zfs_trunc(znode_t *zp, uint64_t end) 1655{ 1656 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1657 vnode_t *vp = ZTOV(zp); 1658 dmu_tx_t *tx; 1659 rl_t *rl; 1660 int error; 1661 sa_bulk_attr_t bulk[2]; 1662 int count = 0; 1663 1664 /* 1665 * We will change zp_size, lock the whole file. 1666 */ 1667 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1668 1669 /* 1670 * Nothing to do if file already at desired length. 1671 */ 1672 if (end >= zp->z_size) { 1673 zfs_range_unlock(rl); 1674 return (0); 1675 } 1676 1677 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1678 if (error) { 1679 zfs_range_unlock(rl); 1680 return (error); 1681 } 1682top: 1683 tx = dmu_tx_create(zfsvfs->z_os); 1684 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1685 zfs_sa_upgrade_txholds(tx, zp); 1686 error = dmu_tx_assign(tx, TXG_NOWAIT); 1687 if (error) { 1688 if (error == ERESTART) { 1689 dmu_tx_wait(tx); 1690 dmu_tx_abort(tx); 1691 goto top; 1692 } 1693 dmu_tx_abort(tx); 1694 zfs_range_unlock(rl); 1695 return (error); 1696 } 1697 1698 zp->z_size = end; 1699 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1700 NULL, &zp->z_size, sizeof (zp->z_size)); 1701 1702 if (end == 0) { 1703 zp->z_pflags &= ~ZFS_SPARSE; 1704 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1705 NULL, &zp->z_pflags, 8); 1706 } 1707 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1708 1709 dmu_tx_commit(tx); 1710 1711 /* 1712 * Clear any mapped pages in the truncated region. This has to 1713 * happen outside of the transaction to avoid the possibility of 1714 * a deadlock with someone trying to push a page that we are 1715 * about to invalidate. 1716 */ 1717 vnode_pager_setsize(vp, end); 1718 1719 zfs_range_unlock(rl); 1720 1721 return (0); 1722} 1723 1724/* 1725 * Free space in a file 1726 * 1727 * IN: zp - znode of file to free data in. 1728 * off - start of range 1729 * len - end of range (0 => EOF) 1730 * flag - current file open mode flags. 1731 * log - TRUE if this action should be logged 1732 * 1733 * RETURN: 0 if success 1734 * error code if failure 1735 */ 1736int 1737zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1738{ 1739 vnode_t *vp = ZTOV(zp); 1740 dmu_tx_t *tx; 1741 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1742 zilog_t *zilog = zfsvfs->z_log; 1743 uint64_t mode; 1744 uint64_t mtime[2], ctime[2]; 1745 sa_bulk_attr_t bulk[3]; 1746 int count = 0; 1747 int error; 1748 1749 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1750 sizeof (mode))) != 0) 1751 return (error); 1752 1753 if (off > zp->z_size) { 1754 error = zfs_extend(zp, off+len); 1755 if (error == 0 && log) 1756 goto log; 1757 else 1758 return (error); 1759 } 1760 1761 /* 1762 * Check for any locks in the region to be freed. 1763 */ 1764 1765 if (MANDLOCK(vp, (mode_t)mode)) { 1766 uint64_t length = (len ? len : zp->z_size - off); 1767 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1768 return (error); 1769 } 1770 1771 if (len == 0) { 1772 error = zfs_trunc(zp, off); 1773 } else { 1774 if ((error = zfs_free_range(zp, off, len)) == 0 && 1775 off + len > zp->z_size) 1776 error = zfs_extend(zp, off+len); 1777 } 1778 if (error || !log) 1779 return (error); 1780log: 1781 tx = dmu_tx_create(zfsvfs->z_os); 1782 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1783 zfs_sa_upgrade_txholds(tx, zp); 1784 error = dmu_tx_assign(tx, TXG_NOWAIT); 1785 if (error) { 1786 if (error == ERESTART) { 1787 dmu_tx_wait(tx); 1788 dmu_tx_abort(tx); 1789 goto log; 1790 } 1791 dmu_tx_abort(tx); 1792 return (error); 1793 } 1794 1795 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1796 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1798 NULL, &zp->z_pflags, 8); 1799 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1800 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1801 ASSERT(error == 0); 1802 1803 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1804 1805 dmu_tx_commit(tx); 1806 return (0); 1807} 1808 1809void 1810zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1811{ 1812 zfsvfs_t zfsvfs; 1813 uint64_t moid, obj, sa_obj, version; 1814 uint64_t sense = ZFS_CASE_SENSITIVE; 1815 uint64_t norm = 0; 1816 nvpair_t *elem; 1817 int error; 1818 int i; 1819 znode_t *rootzp = NULL; 1820 vnode_t vnode; 1821 vattr_t vattr; 1822 znode_t *zp; 1823 zfs_acl_ids_t acl_ids; 1824 1825 /* 1826 * First attempt to create master node. 1827 */ 1828 /* 1829 * In an empty objset, there are no blocks to read and thus 1830 * there can be no i/o errors (which we assert below). 1831 */ 1832 moid = MASTER_NODE_OBJ; 1833 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1834 DMU_OT_NONE, 0, tx); 1835 ASSERT(error == 0); 1836 1837 /* 1838 * Set starting attributes. 1839 */ 1840 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1841 elem = NULL; 1842 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1843 /* For the moment we expect all zpl props to be uint64_ts */ 1844 uint64_t val; 1845 char *name; 1846 1847 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1848 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1849 name = nvpair_name(elem); 1850 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1851 if (val < version) 1852 version = val; 1853 } else { 1854 error = zap_update(os, moid, name, 8, 1, &val, tx); 1855 } 1856 ASSERT(error == 0); 1857 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1858 norm = val; 1859 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1860 sense = val; 1861 } 1862 ASSERT(version != 0); 1863 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1864 1865 /* 1866 * Create zap object used for SA attribute registration 1867 */ 1868 1869 if (version >= ZPL_VERSION_SA) { 1870 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1871 DMU_OT_NONE, 0, tx); 1872 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1873 ASSERT(error == 0); 1874 } else { 1875 sa_obj = 0; 1876 } 1877 /* 1878 * Create a delete queue. 1879 */ 1880 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1881 1882 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1883 ASSERT(error == 0); 1884 1885 /* 1886 * Create root znode. Create minimal znode/vnode/zfsvfs 1887 * to allow zfs_mknode to work. 1888 */ 1889 VATTR_NULL(&vattr); 1890 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1891 vattr.va_type = VDIR; 1892 vattr.va_mode = S_IFDIR|0755; 1893 vattr.va_uid = crgetuid(cr); 1894 vattr.va_gid = crgetgid(cr); 1895 1896 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1897 1898 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1899 zfs_znode_cache_constructor(rootzp, NULL, 0); 1900 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1901 rootzp->z_moved = 0; 1902 rootzp->z_unlinked = 0; 1903 rootzp->z_atime_dirty = 0; 1904 rootzp->z_is_sa = USE_SA(version, os); 1905 1906 vnode.v_type = VDIR; 1907 vnode.v_data = rootzp; 1908 rootzp->z_vnode = &vnode; 1909 1910 zfsvfs.z_os = os; 1911 zfsvfs.z_parent = &zfsvfs; 1912 zfsvfs.z_version = version; 1913 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1914 zfsvfs.z_use_sa = USE_SA(version, os); 1915 zfsvfs.z_norm = norm; 1916 1917 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1918 &zfsvfs.z_attr_table); 1919 1920 ASSERT(error == 0); 1921 1922 /* 1923 * Fold case on file systems that are always or sometimes case 1924 * insensitive. 1925 */ 1926 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1927 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1928 1929 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1930 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1931 offsetof(znode_t, z_link_node)); 1932 1933 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1934 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1935 1936 rootzp->z_zfsvfs = &zfsvfs; 1937 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1938 cr, NULL, &acl_ids)); 1939 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1940 ASSERT3P(zp, ==, rootzp); 1941 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1942 ASSERT(error == 0); 1943 zfs_acl_ids_free(&acl_ids); 1944 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1945 1946 sa_handle_destroy(rootzp->z_sa_hdl); 1947 rootzp->z_vnode = NULL; 1948 kmem_cache_free(znode_cache, rootzp); 1949 1950 /* 1951 * Create shares directory 1952 */ 1953 1954 error = zfs_create_share_dir(&zfsvfs, tx); 1955 1956 ASSERT(error == 0); 1957 1958 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1959 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1960} 1961 1962#endif /* _KERNEL */ 1963 1964static int 1965zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1966{ 1967 uint64_t sa_obj = 0; 1968 int error; 1969 1970 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1971 if (error != 0 && error != ENOENT) 1972 return (error); 1973 1974 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1975 return (error); 1976} 1977 1978static int 1979zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1980 dmu_buf_t **db, void *tag) 1981{ 1982 dmu_object_info_t doi; 1983 int error; 1984 1985 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1986 return (error); 1987 1988 dmu_object_info_from_db(*db, &doi); 1989 if ((doi.doi_bonus_type != DMU_OT_SA && 1990 doi.doi_bonus_type != DMU_OT_ZNODE) || 1991 doi.doi_bonus_type == DMU_OT_ZNODE && 1992 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1993 sa_buf_rele(*db, tag); 1994 return (ENOTSUP); 1995 } 1996 1997 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1998 if (error != 0) { 1999 sa_buf_rele(*db, tag); 2000 return (error); 2001 } 2002 2003 return (0); 2004} 2005 2006void 2007zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2008{ 2009 sa_handle_destroy(hdl); 2010 sa_buf_rele(db, tag); 2011} 2012 2013/* 2014 * Given an object number, return its parent object number and whether 2015 * or not the object is an extended attribute directory. 2016 */ 2017static int 2018zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp, 2019 int *is_xattrdir) 2020{ 2021 uint64_t parent; 2022 uint64_t pflags; 2023 uint64_t mode; 2024 sa_bulk_attr_t bulk[3]; 2025 int count = 0; 2026 int error; 2027 2028 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2029 &parent, sizeof (parent)); 2030 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2031 &pflags, sizeof (pflags)); 2032 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2033 &mode, sizeof (mode)); 2034 2035 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2036 return (error); 2037 2038 *pobjp = parent; 2039 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2040 2041 return (0); 2042} 2043 2044/* 2045 * Given an object number, return some zpl level statistics 2046 */ 2047static int 2048zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2049 zfs_stat_t *sb) 2050{ 2051 sa_bulk_attr_t bulk[4]; 2052 int count = 0; 2053 2054 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2055 &sb->zs_mode, sizeof (sb->zs_mode)); 2056 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2057 &sb->zs_gen, sizeof (sb->zs_gen)); 2058 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2059 &sb->zs_links, sizeof (sb->zs_links)); 2060 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2061 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2062 2063 return (sa_bulk_lookup(hdl, bulk, count)); 2064} 2065 2066static int 2067zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2068 sa_attr_type_t *sa_table, char *buf, int len) 2069{ 2070 sa_handle_t *sa_hdl; 2071 sa_handle_t *prevhdl = NULL; 2072 dmu_buf_t *prevdb = NULL; 2073 dmu_buf_t *sa_db = NULL; 2074 char *path = buf + len - 1; 2075 int error; 2076 2077 *path = '\0'; 2078 sa_hdl = hdl; 2079 2080 for (;;) { 2081 uint64_t pobj; 2082 char component[MAXNAMELEN + 2]; 2083 size_t complen; 2084 int is_xattrdir; 2085 2086 if (prevdb) 2087 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2088 2089 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj, 2090 &is_xattrdir)) != 0) 2091 break; 2092 2093 if (pobj == obj) { 2094 if (path[0] != '/') 2095 *--path = '/'; 2096 break; 2097 } 2098 2099 component[0] = '/'; 2100 if (is_xattrdir) { 2101 (void) sprintf(component + 1, "<xattrdir>"); 2102 } else { 2103 error = zap_value_search(osp, pobj, obj, 2104 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2105 if (error != 0) 2106 break; 2107 } 2108 2109 complen = strlen(component); 2110 path -= complen; 2111 ASSERT(path >= buf); 2112 bcopy(component, path, complen); 2113 obj = pobj; 2114 2115 if (sa_hdl != hdl) { 2116 prevhdl = sa_hdl; 2117 prevdb = sa_db; 2118 } 2119 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2120 if (error != 0) { 2121 sa_hdl = prevhdl; 2122 sa_db = prevdb; 2123 break; 2124 } 2125 } 2126 2127 if (sa_hdl != NULL && sa_hdl != hdl) { 2128 ASSERT(sa_db != NULL); 2129 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2130 } 2131 2132 if (error == 0) 2133 (void) memmove(buf, path, buf + len - path); 2134 2135 return (error); 2136} 2137 2138int 2139zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2140{ 2141 sa_attr_type_t *sa_table; 2142 sa_handle_t *hdl; 2143 dmu_buf_t *db; 2144 int error; 2145 2146 error = zfs_sa_setup(osp, &sa_table); 2147 if (error != 0) 2148 return (error); 2149 2150 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2151 if (error != 0) 2152 return (error); 2153 2154 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2155 2156 zfs_release_sa_handle(hdl, db, FTAG); 2157 return (error); 2158} 2159 2160int 2161zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2162 char *buf, int len) 2163{ 2164 char *path = buf + len - 1; 2165 sa_attr_type_t *sa_table; 2166 sa_handle_t *hdl; 2167 dmu_buf_t *db; 2168 int error; 2169 2170 *path = '\0'; 2171 2172 error = zfs_sa_setup(osp, &sa_table); 2173 if (error != 0) 2174 return (error); 2175 2176 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2177 if (error != 0) 2178 return (error); 2179 2180 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2181 if (error != 0) { 2182 zfs_release_sa_handle(hdl, db, FTAG); 2183 return (error); 2184 } 2185 2186 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2187 2188 zfs_release_sa_handle(hdl, db, FTAG); 2189 return (error); 2190} 2191